Few-shot Learning with Data Enhancement and Transfer Learning for Underwater Target Recognition

Due to the difficulty of acquiring underwater target data, the recognition of small samples of underwater acoustic targets has always been a difficult problem in the field of underwater acoustics. To solve this problem, this paper proposes a few-shot learning method based on data enhancement and transfer learning for underwater target recognition. This paper takes two-dimensional time-frequency spectrum as input, and uses a variety of data enhancement schemes, combined with transfer learning methods to achieve target classification. In terms of experiments, we used hydrophones to collect 3 times data in different sea areas as a data set for verification, including 10 types of targets. The experimental results show that our system could achieve the accuracy of 0.82.